Lecture Notes: 30 Modern FS

ZFS Features #

Copy on Write

• Writes don’t overwrite blocks.
• Instead, the new version of the block is written to a new location.
• Metadata lives in a tree, like in a log-structured FS.
• Every change creates new parent nodes, up to a new root.
• (Draw a persistent tree)
• This doesn’t change the old tree, so corruption can always be dealt with by rolling back to the old root.
• To detect corruption, every node gets a checksum. This also helps detect corruption due to hardware issues.
• This also allows mostly sequential writes.
• Bonus: This gives easy snapshot functionality by pinning old roots.
• Reflink copies

Built-in Multi-Disk Support

• A ZFS filesystem exists on a pool of disks.
• This replaces traditional RAID modes.
• More advantages: e.g. node checksums can be used to detect bad disks.
• Modes: raid-z (1 parity), raid-z2, raid-z3

ZFS History / Licensing #

• Solaris, OpenSolaris
• License issue for Linux
• FreeBSD

BTRFS #

• Basically a rewrite of ZFS for Linux license compatibility.
• Provides good support for duplicate modes but not for parity modes.

Advantages over ZFS:

• Works on stock Linux
• Definitely legal on Linux
• Allows pools to be shrunk
• Broader recognition in some other tools that use reflinks, snapshots, subvolumes on Linux
• Better performance on lower RAM systems.

Advantages of ZFS over BTRFS:

• More mature.
• Better performance with lots of RAM.
• Has stable raid-z parity modes.
• Great on FreeBSD
• Doesn’t get completely borked occasionally because it can’t manage mostly-full disks.

Recommendations:

• If you’re building a big file server, consider ZFS on FreeBSD.
• If you don’t want to worry about filesystems and have backups, ext4 on Linux is fine.
• If you care about your data on Linux, have backups and run a 2+ disk BTRFS pool with “raid1”.

Homework #

Introduce the HW.

Key idea: Two terminals, run the FS in one and try stuff in the other.